The Developer's Note-Taking Problem — and Why Existing Tools Fall Short
The Developer's Note-Taking Problem — and Why Existing Tools Fall Short
There is a quiet frustration shared by a large fraction of working developers: the tools built to capture thoughts and knowledge do not fit the way developers actually work.
This is a discussion about that gap — what causes it, why it persists, and what a better answer looks like.
The shape of developer work
A developer's day is not one long uninterrupted coding session. It is a series of context switches: reading a ticket, grepping a codebase, SSHing into a server, reading logs, writing a fix, reviewing a colleague's PR, debugging a flaky test, then writing up what was learned.
Throughout all of this, there is a constant need to externalise working memory. Notes are not a luxury — they are a cognitive tool. A place to park a hypothesis while you go test it. A running log of what you tried and why it failed. A record of the three-line workaround that fixed a thing six months ago and which you will definitely need again.
The friction in capturing that information matters enormously. A note not taken because the tool was slow or inconvenient is a note lost.
What existing tools get wrong
Electron apps are heavy. An app that takes two seconds to open and uses 300 MB of RAM to display plain text is not a fast tool. It is a slow tool that looks fast because your machine is powerful. Over a day of context switches, that overhead adds up. More importantly, Electron apps are not available over SSH — which is where a significant fraction of developer work actually happens.
Cloud-first tools assume connectivity and trust. Notion, Roam, Confluence — these are excellent collaborative tools for teams that want a shared workspace. They are poor tools for capturing personal working notes, because every note is transmitted to and stored on someone else's server. For notes that contain proprietary code, customer data, security findings, or internal architecture, this is a non-starter.
File-based tools (Obsidian, plain Markdown) are close but not quite. The idea is right — local files, your own storage, full control — but the execution is usually an Electron app that runs a local web server. The editing experience is a web textarea. Full-text search is an index that needs warming up. And there is no answer for "I am SSH'd into a server and need to write something down right now."
Purpose-built developer tools (wikis, Jira, Confluence) are not personal. They are collaborative by design, which means they have process overhead that is exactly wrong for a single developer's scratch notes.
The case for terminal-native
The terminal is the developer's native environment. A tool that lives there does not require a context switch to use. It starts instantly. It works over SSH. It works offline. It imposes no process overhead.
The terminal also has a useful property: it is text-first. There is no temptation to add drag-and-drop kanban boards or animated page transitions. A terminal tool that does notes well is a tool that is fast, searchable, and keyboard-driven. Those are exactly the properties a developer needs.
The case for local storage
A SQLite file is a remarkably good note store. It is a single file. You can back it up with cp. You can put it on a network drive, in Dropbox, or in a git repository. You can query it directly with sqlite3. You can open the same database on Linux, macOS, Windows, Android, and iOS from the same application binary.
Local storage also means local AI. Semantic search and Q&A over your notes does not require an API key or an internet connection. A quantised 135M parameter model and a 384-dimension embedding model fit in the same binary as the rest of the application and run entirely on CPU in a few hundred milliseconds.
What the right tool looks like
A terminal-native, offline-first, SQLite-backed note tool that:
- Starts in under 100ms
- Runs identically over SSH
- Has a real editor (find, replace, undo, selection, image paste)
- Has FTS5 full-text search across all notes
- Has Lua-backed templates for recurring structures
- Has a local AI inference engine for semantic search and Q&A
- Has a code navigation layer for researching codebases
- Has a spreadsheet engine for tabular data and Lua math
- Ships as a single ~2 MB binary with no runtime dependencies
That is Syahi. Download it and try it in your next terminal session.